Skip to contentSkip to conversation

Several perf metrics topdown related fixes

PATCH Superseded [email protected] 2 participants 13 messages May 13, 2022 perf events

v1 of 3 Superseded Latest is v3 →

lore
Share & tools
lore
#0 lore raw

cover Several perf metrics topdown related fixes

From: Kan Liang <[email protected]>

The Patch 1 is a follow-up patch for Ian's ("Fix topdown event weak grouping")[1].

The patch 2 is to fix the perf metrics topdown events in a mixed group. It reuses the function introduced in [1].
Patch 1 & 2 should be on top of [1].

The patch 3 & 4 are to fix other perf metrics topdown related issues. They can be merged separately.

[1]: https://lore.kernel.org/all/[email protected]/

Kan Liang (4):
perf evsel: Fixes topdown events in a weak group for the hybrid
platform
perf stat: Always keep perf metrics topdown events in a group perf parse-events: Support different format of the topdown event name perf parse-events: Move slots event for hybrid platforms too

 tools/perf/arch/x86/util/evlist.c  |  7 ++++---
 tools/perf/arch/x86/util/evsel.c   |  5 +++--
 tools/perf/arch/x86/util/topdown.c | 18 ++++++++++++++++++
 tools/perf/arch/x86/util/topdown.h |  7 +++++++
 tools/perf/builtin-stat.c          |  7 +++++--
 5 files changed, 37 insertions(+), 7 deletions(-)
 create mode 100644 tools/perf/arch/x86/util/topdown.h
Signature · 1 line
2.35.1
#1 lore raw

PATCH 1/4 perf evsel: Fixes topdown events in a weak group for the hybrid platform

From: Kan Liang <[email protected]>

The patch ("perf evlist: Keep topdown counters in weak group") fixes the perf metrics topdown event issue when the topdown events are in a weak group on a non-hybrid platform. However, it doesn't work for the hybrid platform.

$./perf stat -e '{cpu_core/slots/,cpu_core/topdown-bad-spec/, cpu_core/topdown-be-bound/,cpu_core/topdown-fe-bound/, cpu_core/topdown-retiring/,cpu_core/branch-instructions/, cpu_core/branch-misses/,cpu_core/bus-cycles/,cpu_core/cache-misses/, cpu_core/cache-references/,cpu_core/cpu-cycles/,cpu_core/instructions/, cpu_core/mem-loads/,cpu_core/mem-stores/,cpu_core/ref-cycles/, cpu_core/cache-misses/,cpu_core/cache-references/}:W' -a sleep 1

Performance counter stats for 'system wide':

       751,765,068      cpu_core/slots/                                               (84.07%)
   <not supported>      cpu_core/topdown-bad-spec/
   <not supported>      cpu_core/topdown-be-bound/
   <not supported>      cpu_core/topdown-fe-bound/
   <not supported>      cpu_core/topdown-retiring/
        12,398,197      cpu_core/branch-instructions/                                     (84.07%)
         1,054,218      cpu_core/branch-misses/                                       (84.24%)
       539,764,637      cpu_core/bus-cycles/                                          (84.64%)
            14,683      cpu_core/cache-misses/                                        (84.87%)
         7,277,809      cpu_core/cache-references/                                     (77.30%)
       222,299,439      cpu_core/cpu-cycles/                                          (77.28%)
        63,661,714      cpu_core/instructions/                                        (84.85%)
                 0      cpu_core/mem-loads/                                           (77.29%)
        12,271,725      cpu_core/mem-stores/                                          (77.30%)
       542,241,102      cpu_core/ref-cycles/                                          (84.85%)
             8,854      cpu_core/cache-misses/                                        (76.71%)
         7,179,013      cpu_core/cache-references/                                     (76.31%)
       1.003245250 seconds time elapsed

A hybrid platform has a different PMU name for the core PMUs, while the current perf hard code the PMU name "cpu".

The evsel->pmu_name can be used to replace the "cpu" to fix the issue. For a hybrid platform, the pmu_name must be non-NULL. Because there are at least two core PMUs. The PMU has to be specified.
For a non-hybrid platform, the pmu_name may be NULL. Because there is only one core PMU, "cpu". For a NULL pmu_name, we can safely assume that it is a "cpu" PMU.

With the patch,

$perf stat -e '{cpu_core/slots/,cpu_core/topdown-bad-spec/, cpu_core/topdown-be-bound/,cpu_core/topdown-fe-bound/, cpu_core/topdown-retiring/,cpu_core/branch-instructions/, cpu_core/branch-misses/,cpu_core/bus-cycles/,cpu_core/cache-misses/, cpu_core/cache-references/,cpu_core/cpu-cycles/,cpu_core/instructions/, cpu_core/mem-loads/,cpu_core/mem-stores/,cpu_core/ref-cycles/, cpu_core/cache-misses/,cpu_core/cache-references/}:W' -a sleep 1

Performance counter stats for 'system wide':

       766,620,266      cpu_core/slots/                                               (84.06%)
        73,172,129      cpu_core/topdown-bad-spec/ #      9.5% bad speculation         (84.06%)
       193,443,341      cpu_core/topdown-be-bound/ #     25.0% backend bound           (84.06%)
       403,940,929      cpu_core/topdown-fe-bound/ #     52.3% frontend bound          (84.06%)
       102,070,237      cpu_core/topdown-retiring/ #     13.2% retiring                (84.06%)
        12,364,429      cpu_core/branch-instructions/                                     (84.03%)
         1,080,124      cpu_core/branch-misses/                                       (84.24%)
       564,120,383      cpu_core/bus-cycles/                                          (84.65%)
            36,979      cpu_core/cache-misses/                                        (84.86%)
         7,298,094      cpu_core/cache-references/                                     (77.30%)
       227,174,372      cpu_core/cpu-cycles/                                          (77.31%)
        63,886,523      cpu_core/instructions/                                        (84.87%)
                 0      cpu_core/mem-loads/                                           (77.31%)
        12,208,782      cpu_core/mem-stores/                                          (77.31%)
       566,409,738      cpu_core/ref-cycles/                                          (84.87%)
            23,118      cpu_core/cache-misses/                                        (76.71%)
         7,212,602      cpu_core/cache-references/                                     (76.29%)
       1.003228667 seconds time elapsed
---
 tools/perf/arch/x86/util/evsel.c | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)
tools/perf/arch/x86/util/evsel.c +3 -2
index 00cb4466b4ca..24510bcb4bf4 100644
@@ -33,8 +33,9 @@ void arch_evsel__fixup_new_cycles(struct perf_event_attr *attr)
 
 bool arch_evsel__must_be_in_group(const struct evsel *evsel)
 {
-	if ((evsel->pmu_name && strcmp(evsel->pmu_name, "cpu")) ||
-	    !pmu_have_event("cpu", "slots"))
+	const char *pmu_name = evsel->pmu_name ? evsel->pmu_name : "cpu";
+
+	if (!pmu_have_event(pmu_name, "slots"))
 		return false;
 
 	return evsel->name &&
Signature · 1 line
2.35.1
reviewer
#2 lore raw

Re: [PATCH 1/4] perf evsel: Fixes topdown events in a weak group for the hybrid platform

Quoted 90 lines · On Fri, May 13, 2022 at 8:16 AM <[email protected]> wrote:

From: Kan Liang <[email protected]>

The patch ("perf evlist: Keep topdown counters in weak group") fixes the perf metrics topdown event issue when the topdown events are in a weak group on a non-hybrid platform. However, it doesn't work for the hybrid platform.

$./perf stat -e '{cpu_core/slots/,cpu_core/topdown-bad-spec/, cpu_core/topdown-be-bound/,cpu_core/topdown-fe-bound/, cpu_core/topdown-retiring/,cpu_core/branch-instructions/, cpu_core/branch-misses/,cpu_core/bus-cycles/,cpu_core/cache-misses/, cpu_core/cache-references/,cpu_core/cpu-cycles/,cpu_core/instructions/, cpu_core/mem-loads/,cpu_core/mem-stores/,cpu_core/ref-cycles/, cpu_core/cache-misses/,cpu_core/cache-references/}:W' -a sleep 1

Performance counter stats for 'system wide':

       751,765,068      cpu_core/slots/                                               (84.07%)
   <not supported>      cpu_core/topdown-bad-spec/
   <not supported>      cpu_core/topdown-be-bound/
   <not supported>      cpu_core/topdown-fe-bound/
   <not supported>      cpu_core/topdown-retiring/
        12,398,197      cpu_core/branch-instructions/                                     (84.07%)
         1,054,218      cpu_core/branch-misses/                                       (84.24%)
       539,764,637      cpu_core/bus-cycles/                                          (84.64%)
            14,683      cpu_core/cache-misses/                                        (84.87%)
         7,277,809      cpu_core/cache-references/                                     (77.30%)
       222,299,439      cpu_core/cpu-cycles/                                          (77.28%)
        63,661,714      cpu_core/instructions/                                        (84.85%)
                 0      cpu_core/mem-loads/                                           (77.29%)
        12,271,725      cpu_core/mem-stores/                                          (77.30%)
       542,241,102      cpu_core/ref-cycles/                                          (84.85%)
             8,854      cpu_core/cache-misses/                                        (76.71%)
         7,179,013      cpu_core/cache-references/                                     (76.31%)
       1.003245250 seconds time elapsed

A hybrid platform has a different PMU name for the core PMUs, while the current perf hard code the PMU name "cpu".

The evsel->pmu_name can be used to replace the "cpu" to fix the issue. For a hybrid platform, the pmu_name must be non-NULL. Because there are at least two core PMUs. The PMU has to be specified.
For a non-hybrid platform, the pmu_name may be NULL. Because there is only one core PMU, "cpu". For a NULL pmu_name, we can safely assume that it is a "cpu" PMU.

With the patch,

$perf stat -e '{cpu_core/slots/,cpu_core/topdown-bad-spec/, cpu_core/topdown-be-bound/,cpu_core/topdown-fe-bound/, cpu_core/topdown-retiring/,cpu_core/branch-instructions/, cpu_core/branch-misses/,cpu_core/bus-cycles/,cpu_core/cache-misses/, cpu_core/cache-references/,cpu_core/cpu-cycles/,cpu_core/instructions/, cpu_core/mem-loads/,cpu_core/mem-stores/,cpu_core/ref-cycles/, cpu_core/cache-misses/,cpu_core/cache-references/}:W' -a sleep 1

Performance counter stats for 'system wide':

       766,620,266      cpu_core/slots/                                               (84.06%)
        73,172,129      cpu_core/topdown-bad-spec/ #      9.5% bad speculation         (84.06%)
       193,443,341      cpu_core/topdown-be-bound/ #     25.0% backend bound           (84.06%)
       403,940,929      cpu_core/topdown-fe-bound/ #     52.3% frontend bound          (84.06%)
       102,070,237      cpu_core/topdown-retiring/ #     13.2% retiring                (84.06%)
        12,364,429      cpu_core/branch-instructions/                                     (84.03%)
         1,080,124      cpu_core/branch-misses/                                       (84.24%)
       564,120,383      cpu_core/bus-cycles/                                          (84.65%)
            36,979      cpu_core/cache-misses/                                        (84.86%)
         7,298,094      cpu_core/cache-references/                                     (77.30%)
       227,174,372      cpu_core/cpu-cycles/                                          (77.31%)
        63,886,523      cpu_core/instructions/                                        (84.87%)
                 0      cpu_core/mem-loads/                                           (77.31%)
        12,208,782      cpu_core/mem-stores/                                          (77.31%)
       566,409,738      cpu_core/ref-cycles/                                          (84.87%)
            23,118      cpu_core/cache-misses/                                        (76.71%)
         7,212,602      cpu_core/cache-references/                                     (76.29%)
       1.003228667 seconds time elapsed
---
 tools/perf/arch/x86/util/evsel.c | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)
tools/perf/arch/x86/util/evsel.c +0 -0
index 00cb4466b4ca..24510bcb4bf4 100644

tools/perf/arch/x86/util/evsel.c void arch_evsel__fixup_new_cycles(struct perf_event_attr *attr) · line 38

Quoted 8 lines
tools/perf/arch/x86/util/evsel.c +3 -2
@@ -33,8 +33,9 @@ void arch_evsel__fixup_new_cycles(struct perf_event_attr *attr)
 
 bool arch_evsel__must_be_in_group(const struct evsel *evsel)
 {
-       if ((evsel->pmu_name && strcmp(evsel->pmu_name, "cpu")) ||
-           !pmu_have_event("cpu", "slots"))
+       const char *pmu_name = evsel->pmu_name ? evsel->pmu_name : "cpu";
+
+       if (!pmu_have_event(pmu_name, "slots"))

Playing devil's advocate, if I have a PMU for my network accelerator and it has an event called "slots" then this test will also be true.

The property that is being tested here is "does this CPU have topdown events" and so allowing any PMU removes the "does this CPU" part of the equation. I think ideally we'd have an arch functions something like:

bool arch_pmu__has_intel_topdown_events(void)
{
static bool has_topdown_events = pmu_have_event("cpu", "slots") ||
pmu_have_event("cpu_core", "slots");

return has_topdown_events;
}

bool arch_pmu__supports_intel_topdown_events(const char *pmu_name) {
if (!pmu_name)

    return false;

return arch_pmu__has_intel_topdown_events() && (!strncmp(pmu_name,
"cpu") || !strncmp(pmu_name, "cpu_core"));
}

bool arch_evsel__is_intel_topdown_event(struct evsel *evsel) {
if (!arch_pmu__supports_intel_topdown_events(evsel->pmu))

    return false;

return strcasestr(evsel->name, "slots") || strcasestr(evsel->name, "topdown");
}

This then gives us:

bool arch_evsel__must_be_in_group(const struct evsel *evsel) {
return arch_evsel__is_intel_topdown_event(evsel);
}

These functions can then be reused for the arch_evlist topdown code, etc. What I don't see in these functions is use of any hybrid abstraction and so it isn't clear to me how with hybrid something like this would be plumbed in.

Thanks,
Ian

Quoted 6 lines
tools/perf/arch/x86/util/evsel.c +0 -1
                return false;
 
        return evsel->name &&
--

2.35.1

author
#3 lore raw

Re: [PATCH 1/4] perf evsel: Fixes topdown events in a weak group for the hybrid platform

Quoted 92 lines · Ian Rogers
Quoted 91 lines · On Fri, May 13, 2022 at 8:16 AM <[email protected]> wrote:

From: Kan Liang <[email protected]>

The patch ("perf evlist: Keep topdown counters in weak group") fixes the perf metrics topdown event issue when the topdown events are in a weak group on a non-hybrid platform. However, it doesn't work for the hybrid platform.

$./perf stat -e '{cpu_core/slots/,cpu_core/topdown-bad-spec/, cpu_core/topdown-be-bound/,cpu_core/topdown-fe-bound/, cpu_core/topdown-retiring/,cpu_core/branch-instructions/, cpu_core/branch-misses/,cpu_core/bus-cycles/,cpu_core/cache-misses/, cpu_core/cache-references/,cpu_core/cpu-cycles/,cpu_core/instructions/, cpu_core/mem-loads/,cpu_core/mem-stores/,cpu_core/ref-cycles/, cpu_core/cache-misses/,cpu_core/cache-references/}:W' -a sleep 1

Performance counter stats for 'system wide':

        751,765,068      cpu_core/slots/                                               (84.07%)
    <not supported>      cpu_core/topdown-bad-spec/
    <not supported>      cpu_core/topdown-be-bound/
    <not supported>      cpu_core/topdown-fe-bound/
    <not supported>      cpu_core/topdown-retiring/
         12,398,197      cpu_core/branch-instructions/                                     (84.07%)
          1,054,218      cpu_core/branch-misses/                                       (84.24%)
        539,764,637      cpu_core/bus-cycles/                                          (84.64%)
             14,683      cpu_core/cache-misses/                                        (84.87%)
          7,277,809      cpu_core/cache-references/                                     (77.30%)
        222,299,439      cpu_core/cpu-cycles/                                          (77.28%)
         63,661,714      cpu_core/instructions/                                        (84.85%)
                  0      cpu_core/mem-loads/                                           (77.29%)
         12,271,725      cpu_core/mem-stores/                                          (77.30%)
        542,241,102      cpu_core/ref-cycles/                                          (84.85%)
              8,854      cpu_core/cache-misses/                                        (76.71%)
          7,179,013      cpu_core/cache-references/                                     (76.31%)
        1.003245250 seconds time elapsed

A hybrid platform has a different PMU name for the core PMUs, while the current perf hard code the PMU name "cpu".

The evsel->pmu_name can be used to replace the "cpu" to fix the issue. For a hybrid platform, the pmu_name must be non-NULL. Because there are at least two core PMUs. The PMU has to be specified.
For a non-hybrid platform, the pmu_name may be NULL. Because there is only one core PMU, "cpu". For a NULL pmu_name, we can safely assume that it is a "cpu" PMU.

With the patch,

$perf stat -e '{cpu_core/slots/,cpu_core/topdown-bad-spec/, cpu_core/topdown-be-bound/,cpu_core/topdown-fe-bound/, cpu_core/topdown-retiring/,cpu_core/branch-instructions/, cpu_core/branch-misses/,cpu_core/bus-cycles/,cpu_core/cache-misses/, cpu_core/cache-references/,cpu_core/cpu-cycles/,cpu_core/instructions/, cpu_core/mem-loads/,cpu_core/mem-stores/,cpu_core/ref-cycles/, cpu_core/cache-misses/,cpu_core/cache-references/}:W' -a sleep 1

Performance counter stats for 'system wide':

        766,620,266      cpu_core/slots/                                               (84.06%)
         73,172,129      cpu_core/topdown-bad-spec/ #      9.5% bad speculation         (84.06%)
        193,443,341      cpu_core/topdown-be-bound/ #     25.0% backend bound           (84.06%)
        403,940,929      cpu_core/topdown-fe-bound/ #     52.3% frontend bound          (84.06%)
        102,070,237      cpu_core/topdown-retiring/ #     13.2% retiring                (84.06%)
         12,364,429      cpu_core/branch-instructions/                                     (84.03%)
          1,080,124      cpu_core/branch-misses/                                       (84.24%)
        564,120,383      cpu_core/bus-cycles/                                          (84.65%)
             36,979      cpu_core/cache-misses/                                        (84.86%)
          7,298,094      cpu_core/cache-references/                                     (77.30%)
        227,174,372      cpu_core/cpu-cycles/                                          (77.31%)
         63,886,523      cpu_core/instructions/                                        (84.87%)
                  0      cpu_core/mem-loads/                                           (77.31%)
         12,208,782      cpu_core/mem-stores/                                          (77.31%)
        566,409,738      cpu_core/ref-cycles/                                          (84.87%)
             23,118      cpu_core/cache-misses/                                        (76.71%)
          7,212,602      cpu_core/cache-references/                                     (76.29%)
        1.003228667 seconds time elapsed
---
  tools/perf/arch/x86/util/evsel.c | 5 +++--
  1 file changed, 3 insertions(+), 2 deletions(-)
tools/perf/arch/x86/util/evsel.c +0 -0
index 00cb4466b4ca..24510bcb4bf4 100644
@@ -33,8 +33,9 @@ void arch_evsel__fixup_new_cycles(struct perf_event_attr *attr)
 
Quoted 11 lines

tools/perf/arch/x86/util/evsel.c void arch_evsel__fixup_new_cycles(struct perf_event_attr *attr) · line 38

Quoted 7 lines
tools/perf/arch/x86/util/evsel.c +3 -2
  bool arch_evsel__must_be_in_group(const struct evsel *evsel)
  {
-       if ((evsel->pmu_name && strcmp(evsel->pmu_name, "cpu")) ||
-           !pmu_have_event("cpu", "slots"))
+       const char *pmu_name = evsel->pmu_name ? evsel->pmu_name : "cpu";
+
+       if (!pmu_have_event(pmu_name, "slots"))

Playing devil's advocate, if I have a PMU for my network accelerator and it has an event called "slots" then this test will also be true.

IIRC, the pmu_have_event should only check the event which is exposed by the kernel. It's very unlikely that another PMU expose the exact same name.

If you still worry about it, I think we can check the PMU type PERF_TYPE_RAW here, which is reserved for the core PMU. Others cannot use it.

It looks like arch_evsel__must_be_in_group() is the only user for the evsel__sys_has_perf_metrics() for now, so I make it static.

The other pmu_have_event("cpu", "slots") is in evlist.c. topdown_sys_has_perf_metrics() in patch 4 should be used to replace it. I think Zhengjun will post patches for the changes for the evlist.c

tools/perf/arch/x86/util/evsel.c +11 -2
index 24510bcb4bf4..a4714174e30f 100644
@@ -31,11 +31,20 @@ void arch_evsel__fixup_new_cycles(struct perf_event_attr *attr)
        free(env.cpuid);
 }
 
-bool arch_evsel__must_be_in_group(const struct evsel *evsel)
+static bool evsel__sys_has_perf_metrics(const struct evsel *evsel)
 {
        const char *pmu_name = evsel->pmu_name ? evsel->pmu_name : "cpu";
 
-       if (!pmu_have_event(pmu_name, "slots"))
+       if ((evsel->core.attr.type == PERF_TYPE_RAW) &&
+           pmu_have_event(pmu_name, "slots"))
+               return true;
+
+       return false;
+}
+
+bool arch_evsel__must_be_in_group(const struct evsel *evsel)
+{
+       if (!evsel__sys_has_perf_metrics(evsel))
                return false;
 
        return evsel->name &&

Thanks,
Kan

Quoted 50 lines

The property that is being tested here is "does this CPU have topdown events" and so allowing any PMU removes the "does this CPU" part of the equation. I think ideally we'd have an arch functions something like:

bool arch_pmu__has_intel_topdown_events(void)

{
   static bool has_topdown_events = pmu_have_event("cpu", "slots") ||

pmu_have_event("cpu_core", "slots");

   return has_topdown_events;
}

bool arch_pmu__supports_intel_topdown_events(const char *pmu_name)

{
   if (!pmu_name)
     return false;
   return arch_pmu__has_intel_topdown_events() && (!strncmp(pmu_name,

"cpu") || !strncmp(pmu_name, "cpu_core"));
}

bool arch_evsel__is_intel_topdown_event(struct evsel *evsel)

{
   if (!arch_pmu__supports_intel_topdown_events(evsel->pmu))
     return false;

   return strcasestr(evsel->name, "slots") || strcasestr(evsel->name, "topdown");
}

This then gives us:

bool arch_evsel__must_be_in_group(const struct evsel *evsel)

{
   return  arch_evsel__is_intel_topdown_event(evsel);
}

These functions can then be reused for the arch_evlist topdown code, etc. What I don't see in these functions is use of any hybrid abstraction and so it isn't clear to me how with hybrid something like this would be plumbed in.

Thanks,
Ian

Quoted 6 lines
tools/perf/arch/x86/util/evsel.c +0 -1
                 return false;
 
         return evsel->name &&
--

2.35.1

reviewer
#4 lore raw

Re: [PATCH 1/4] perf evsel: Fixes topdown events in a weak group for the hybrid platform

Quoted 108 lines · Kan
Quoted 92 lines · Ian Rogers
Quoted 91 lines · On Fri, May 13, 2022 at 8:16 AM <[email protected]> wrote:

From: Kan Liang <[email protected]>

The patch ("perf evlist: Keep topdown counters in weak group") fixes the perf metrics topdown event issue when the topdown events are in a weak group on a non-hybrid platform. However, it doesn't work for the hybrid platform.

$./perf stat -e '{cpu_core/slots/,cpu_core/topdown-bad-spec/, cpu_core/topdown-be-bound/,cpu_core/topdown-fe-bound/, cpu_core/topdown-retiring/,cpu_core/branch-instructions/, cpu_core/branch-misses/,cpu_core/bus-cycles/,cpu_core/cache-misses/, cpu_core/cache-references/,cpu_core/cpu-cycles/,cpu_core/instructions/, cpu_core/mem-loads/,cpu_core/mem-stores/,cpu_core/ref-cycles/, cpu_core/cache-misses/,cpu_core/cache-references/}:W' -a sleep 1

Performance counter stats for 'system wide':

        751,765,068      cpu_core/slots/                                               (84.07%)
    <not supported>      cpu_core/topdown-bad-spec/
    <not supported>      cpu_core/topdown-be-bound/
    <not supported>      cpu_core/topdown-fe-bound/
    <not supported>      cpu_core/topdown-retiring/
         12,398,197      cpu_core/branch-instructions/                                     (84.07%)
          1,054,218      cpu_core/branch-misses/                                       (84.24%)
        539,764,637      cpu_core/bus-cycles/                                          (84.64%)
             14,683      cpu_core/cache-misses/                                        (84.87%)
          7,277,809      cpu_core/cache-references/                                     (77.30%)
        222,299,439      cpu_core/cpu-cycles/                                          (77.28%)
         63,661,714      cpu_core/instructions/                                        (84.85%)
                  0      cpu_core/mem-loads/                                           (77.29%)
         12,271,725      cpu_core/mem-stores/                                          (77.30%)
        542,241,102      cpu_core/ref-cycles/                                          (84.85%)
              8,854      cpu_core/cache-misses/                                        (76.71%)
          7,179,013      cpu_core/cache-references/                                     (76.31%)
        1.003245250 seconds time elapsed

A hybrid platform has a different PMU name for the core PMUs, while the current perf hard code the PMU name "cpu".

The evsel->pmu_name can be used to replace the "cpu" to fix the issue. For a hybrid platform, the pmu_name must be non-NULL. Because there are at least two core PMUs. The PMU has to be specified.
For a non-hybrid platform, the pmu_name may be NULL. Because there is only one core PMU, "cpu". For a NULL pmu_name, we can safely assume that it is a "cpu" PMU.

With the patch,

$perf stat -e '{cpu_core/slots/,cpu_core/topdown-bad-spec/, cpu_core/topdown-be-bound/,cpu_core/topdown-fe-bound/, cpu_core/topdown-retiring/,cpu_core/branch-instructions/, cpu_core/branch-misses/,cpu_core/bus-cycles/,cpu_core/cache-misses/, cpu_core/cache-references/,cpu_core/cpu-cycles/,cpu_core/instructions/, cpu_core/mem-loads/,cpu_core/mem-stores/,cpu_core/ref-cycles/, cpu_core/cache-misses/,cpu_core/cache-references/}:W' -a sleep 1

Performance counter stats for 'system wide':

        766,620,266      cpu_core/slots/                                               (84.06%)
         73,172,129      cpu_core/topdown-bad-spec/ #      9.5% bad speculation         (84.06%)
        193,443,341      cpu_core/topdown-be-bound/ #     25.0% backend bound           (84.06%)
        403,940,929      cpu_core/topdown-fe-bound/ #     52.3% frontend bound          (84.06%)
        102,070,237      cpu_core/topdown-retiring/ #     13.2% retiring                (84.06%)
         12,364,429      cpu_core/branch-instructions/                                     (84.03%)
          1,080,124      cpu_core/branch-misses/                                       (84.24%)
        564,120,383      cpu_core/bus-cycles/                                          (84.65%)
             36,979      cpu_core/cache-misses/                                        (84.86%)
          7,298,094      cpu_core/cache-references/                                     (77.30%)
        227,174,372      cpu_core/cpu-cycles/                                          (77.31%)
         63,886,523      cpu_core/instructions/                                        (84.87%)
                  0      cpu_core/mem-loads/                                           (77.31%)
         12,208,782      cpu_core/mem-stores/                                          (77.31%)
        566,409,738      cpu_core/ref-cycles/                                          (84.87%)
             23,118      cpu_core/cache-misses/                                        (76.71%)
          7,212,602      cpu_core/cache-references/                                     (76.29%)
        1.003228667 seconds time elapsed
---
  tools/perf/arch/x86/util/evsel.c | 5 +++--
  1 file changed, 3 insertions(+), 2 deletions(-)
tools/perf/arch/x86/util/evsel.c +0 -0
index 00cb4466b4ca..24510bcb4bf4 100644
@@ -33,8 +33,9 @@ void arch_evsel__fixup_new_cycles(struct perf_event_attr *attr)
 
Quoted 11 lines

tools/perf/arch/x86/util/evsel.c void arch_evsel__fixup_new_cycles(struct perf_event_attr *attr) · line 38

Quoted 7 lines
tools/perf/arch/x86/util/evsel.c +3 -2
  bool arch_evsel__must_be_in_group(const struct evsel *evsel)
  {
-       if ((evsel->pmu_name && strcmp(evsel->pmu_name, "cpu")) ||
-           !pmu_have_event("cpu", "slots"))
+       const char *pmu_name = evsel->pmu_name ? evsel->pmu_name : "cpu";
+
+       if (!pmu_have_event(pmu_name, "slots"))

Playing devil's advocate, if I have a PMU for my network accelerator and it has an event called "slots" then this test will also be true.

Quoted 6 lines

IIRC, the pmu_have_event should only check the event which is exposed by the kernel. It's very unlikely that another PMU expose the exact same name.

If you still worry about it, I think we can check the PMU type PERF_TYPE_RAW here, which is reserved for the core PMU. Others cannot use it.

That's cool, this isn't documented behavior though:
https://git.kernel.org/pub/scm/linux/kernel/git/acme/linux.git/tree/tools/include/uapi/linux/perf_event.h?h=perf/core#n34
and PERF_TYPE_HARDWARE wouldn't seem a wholly unreasonable type. It kind of feels like depending on a quirk, and so we should bury the quirk in a helper function and document it :-)

Quoted 6 lines

It looks like arch_evsel__must_be_in_group() is the only user for the evsel__sys_has_perf_metrics() for now, so I make it static.

The other pmu_have_event("cpu", "slots") is in evlist.c. topdown_sys_has_perf_metrics() in patch 4 should be used to replace it. I think Zhengjun will post patches for the changes for the evlist.c

Ok, is Zhengjun putting his changes on top of this and fixing up the APIs or is he waiting on these changes landing? Let me know how to help. I'm guessing landing my changes is the first step.

Thanks,
Ian

Quoted 83 lines
tools/perf/arch/x86/util/evsel.c +11 -2
index 24510bcb4bf4..a4714174e30f 100644
@@ -31,11 +31,20 @@ void arch_evsel__fixup_new_cycles(struct
 perf_event_attr *attr)
         free(env.cpuid);
  }
 
-bool arch_evsel__must_be_in_group(const struct evsel *evsel)
+static bool evsel__sys_has_perf_metrics(const struct evsel *evsel)
  {
         const char *pmu_name = evsel->pmu_name ? evsel->pmu_name : "cpu";
 
-       if (!pmu_have_event(pmu_name, "slots"))
+       if ((evsel->core.attr.type == PERF_TYPE_RAW) &&
+           pmu_have_event(pmu_name, "slots"))
+               return true;
+
+       return false;
+}
+
+bool arch_evsel__must_be_in_group(const struct evsel *evsel)
+{
+       if (!evsel__sys_has_perf_metrics(evsel))
                 return false;
 
         return evsel->name &&

Thanks,
Kan

Quoted 50 lines

The property that is being tested here is "does this CPU have topdown events" and so allowing any PMU removes the "does this CPU" part of the equation. I think ideally we'd have an arch functions something like:

bool arch_pmu__has_intel_topdown_events(void)

{
   static bool has_topdown_events = pmu_have_event("cpu", "slots") ||

pmu_have_event("cpu_core", "slots");

   return has_topdown_events;
}

bool arch_pmu__supports_intel_topdown_events(const char *pmu_name)

{
   if (!pmu_name)
     return false;
   return arch_pmu__has_intel_topdown_events() && (!strncmp(pmu_name,

"cpu") || !strncmp(pmu_name, "cpu_core"));
}

bool arch_evsel__is_intel_topdown_event(struct evsel *evsel)

{
   if (!arch_pmu__supports_intel_topdown_events(evsel->pmu))
     return false;

   return strcasestr(evsel->name, "slots") || strcasestr(evsel->name, "topdown");
}

This then gives us:

bool arch_evsel__must_be_in_group(const struct evsel *evsel)

{
   return  arch_evsel__is_intel_topdown_event(evsel);
}

These functions can then be reused for the arch_evlist topdown code, etc. What I don't see in these functions is use of any hybrid abstraction and so it isn't clear to me how with hybrid something like this would be plumbed in.

Thanks,
Ian

Quoted 6 lines
tools/perf/arch/x86/util/evsel.c +0 -1
                 return false;
 
         return evsel->name &&
--

2.35.1

author
#5 lore raw

Re: [PATCH 1/4] perf evsel: Fixes topdown events in a weak group for the hybrid platform

Quoted 116 lines · Ian Rogers
Quoted 108 lines · Kan
Quoted 92 lines · Ian Rogers
Quoted 91 lines · On Fri, May 13, 2022 at 8:16 AM <[email protected]> wrote:

From: Kan Liang <[email protected]>

The patch ("perf evlist: Keep topdown counters in weak group") fixes the perf metrics topdown event issue when the topdown events are in a weak group on a non-hybrid platform. However, it doesn't work for the hybrid platform.

$./perf stat -e '{cpu_core/slots/,cpu_core/topdown-bad-spec/, cpu_core/topdown-be-bound/,cpu_core/topdown-fe-bound/, cpu_core/topdown-retiring/,cpu_core/branch-instructions/, cpu_core/branch-misses/,cpu_core/bus-cycles/,cpu_core/cache-misses/, cpu_core/cache-references/,cpu_core/cpu-cycles/,cpu_core/instructions/, cpu_core/mem-loads/,cpu_core/mem-stores/,cpu_core/ref-cycles/, cpu_core/cache-misses/,cpu_core/cache-references/}:W' -a sleep 1

Performance counter stats for 'system wide':

         751,765,068      cpu_core/slots/                                               (84.07%)
     <not supported>      cpu_core/topdown-bad-spec/
     <not supported>      cpu_core/topdown-be-bound/
     <not supported>      cpu_core/topdown-fe-bound/
     <not supported>      cpu_core/topdown-retiring/
          12,398,197      cpu_core/branch-instructions/                                     (84.07%)
           1,054,218      cpu_core/branch-misses/                                       (84.24%)
         539,764,637      cpu_core/bus-cycles/                                          (84.64%)
              14,683      cpu_core/cache-misses/                                        (84.87%)
           7,277,809      cpu_core/cache-references/                                     (77.30%)
         222,299,439      cpu_core/cpu-cycles/                                          (77.28%)
          63,661,714      cpu_core/instructions/                                        (84.85%)
                   0      cpu_core/mem-loads/                                           (77.29%)
          12,271,725      cpu_core/mem-stores/                                          (77.30%)
         542,241,102      cpu_core/ref-cycles/                                          (84.85%)
               8,854      cpu_core/cache-misses/                                        (76.71%)
           7,179,013      cpu_core/cache-references/                                     (76.31%)
         1.003245250 seconds time elapsed

A hybrid platform has a different PMU name for the core PMUs, while the current perf hard code the PMU name "cpu".

The evsel->pmu_name can be used to replace the "cpu" to fix the issue. For a hybrid platform, the pmu_name must be non-NULL. Because there are at least two core PMUs. The PMU has to be specified.
For a non-hybrid platform, the pmu_name may be NULL. Because there is only one core PMU, "cpu". For a NULL pmu_name, we can safely assume that it is a "cpu" PMU.

With the patch,

$perf stat -e '{cpu_core/slots/,cpu_core/topdown-bad-spec/, cpu_core/topdown-be-bound/,cpu_core/topdown-fe-bound/, cpu_core/topdown-retiring/,cpu_core/branch-instructions/, cpu_core/branch-misses/,cpu_core/bus-cycles/,cpu_core/cache-misses/, cpu_core/cache-references/,cpu_core/cpu-cycles/,cpu_core/instructions/, cpu_core/mem-loads/,cpu_core/mem-stores/,cpu_core/ref-cycles/, cpu_core/cache-misses/,cpu_core/cache-references/}:W' -a sleep 1

   Performance counter stats for 'system wide':

         766,620,266      cpu_core/slots/                                               (84.06%)
          73,172,129      cpu_core/topdown-bad-spec/ #      9.5% bad speculation         (84.06%)
         193,443,341      cpu_core/topdown-be-bound/ #     25.0% backend bound           (84.06%)
         403,940,929      cpu_core/topdown-fe-bound/ #     52.3% frontend bound          (84.06%)
         102,070,237      cpu_core/topdown-retiring/ #     13.2% retiring                (84.06%)
          12,364,429      cpu_core/branch-instructions/                                     (84.03%)
           1,080,124      cpu_core/branch-misses/                                       (84.24%)
         564,120,383      cpu_core/bus-cycles/                                          (84.65%)
              36,979      cpu_core/cache-misses/                                        (84.86%)
           7,298,094      cpu_core/cache-references/                                     (77.30%)
         227,174,372      cpu_core/cpu-cycles/                                          (77.31%)
          63,886,523      cpu_core/instructions/                                        (84.87%)
                   0      cpu_core/mem-loads/                                           (77.31%)
          12,208,782      cpu_core/mem-stores/                                          (77.31%)
         566,409,738      cpu_core/ref-cycles/                                          (84.87%)
              23,118      cpu_core/cache-misses/                                        (76.71%)
           7,212,602      cpu_core/cache-references/                                     (76.29%)

         1.003228667 seconds time elapsed
---
   tools/perf/arch/x86/util/evsel.c | 5 +++--
   1 file changed, 3 insertions(+), 2 deletions(-)
tools/perf/arch/x86/util/evsel.c +0 -0
index 00cb4466b4ca..24510bcb4bf4 100644
@@ -33,8 +33,9 @@ void arch_evsel__fixup_new_cycles(struct perf_event_attr *attr)
 
Quoted 11 lines

tools/perf/arch/x86/util/evsel.c void arch_evsel__fixup_new_cycles(struct perf_event_attr *attr) · line 38

Quoted 7 lines
tools/perf/arch/x86/util/evsel.c +3 -2
   bool arch_evsel__must_be_in_group(const struct evsel *evsel)
   {
-       if ((evsel->pmu_name && strcmp(evsel->pmu_name, "cpu")) ||
-           !pmu_have_event("cpu", "slots"))
+       const char *pmu_name = evsel->pmu_name ? evsel->pmu_name : "cpu";
+
+       if (!pmu_have_event(pmu_name, "slots"))

Playing devil's advocate, if I have a PMU for my network accelerator and it has an event called "slots" then this test will also be true.

Quoted 6 lines

IIRC, the pmu_have_event should only check the event which is exposed by the kernel. It's very unlikely that another PMU expose the exact same name.

If you still worry about it, I think we can check the PMU type PERF_TYPE_RAW here, which is reserved for the core PMU. Others cannot use it.

Quoted 5 lines

That's cool, this isn't documented behavior though:
https://git.kernel.org/pub/scm/linux/kernel/git/acme/linux.git/tree/tools/include/uapi/linux/perf_event.h?h=perf/core#n34
and PERF_TYPE_HARDWARE wouldn't seem a wholly unreasonable type. It kind of feels like depending on a quirk, and so we should bury the quirk in a helper function and document it :-)

The PERF_TYPE_HARDWARE and PERF_TYPE_HW_CACHE are aliases for the PERF_TYPE_RAW. The PERF_TYPE_HARDWARE is used for the 10 common hardware events and The PERF_TYPE_HW_CACHE is used for the hardware cache events. Other core events (not include the atom core events in a hybrid machine) should have the PERF_TYPE_RAW type.

Since the perf metrics is a big core only feature, checking both the PERF_TYPE_RAW type and the slots event should be good enough here.

I will add more comments in V2.

Quoted 11 lines
Quoted 7 lines

It looks like arch_evsel__must_be_in_group() is the only user for the evsel__sys_has_perf_metrics() for now, so I make it static.

The other pmu_have_event("cpu", "slots") is in evlist.c. topdown_sys_has_perf_metrics() in patch 4 should be used to replace it. I think Zhengjun will post patches for the changes for the evlist.c

Ok, is Zhengjun putting his changes on top of this and fixing up the APIs or is he waiting on these changes landing? Let me know how to help. I'm guessing landing my changes is the first step.

Right. Your changes is the first step. Then this patch set. Zhengjun's will be on top of us.

Thanks,
Kan

Quoted 87 lines

Thanks,
Ian

Quoted 83 lines
tools/perf/arch/x86/util/evsel.c +11 -2
index 24510bcb4bf4..a4714174e30f 100644
@@ -31,11 +31,20 @@ void arch_evsel__fixup_new_cycles(struct
 perf_event_attr *attr)
          free(env.cpuid);
   }
 
-bool arch_evsel__must_be_in_group(const struct evsel *evsel)
+static bool evsel__sys_has_perf_metrics(const struct evsel *evsel)
   {
          const char *pmu_name = evsel->pmu_name ? evsel->pmu_name : "cpu";
 
-       if (!pmu_have_event(pmu_name, "slots"))
+       if ((evsel->core.attr.type == PERF_TYPE_RAW) &&
+           pmu_have_event(pmu_name, "slots"))
+               return true;
+
+       return false;
+}
+
+bool arch_evsel__must_be_in_group(const struct evsel *evsel)
+{
+       if (!evsel__sys_has_perf_metrics(evsel))
                  return false;
 
          return evsel->name &&

Thanks,
Kan

Quoted 50 lines

The property that is being tested here is "does this CPU have topdown events" and so allowing any PMU removes the "does this CPU" part of the equation. I think ideally we'd have an arch functions something like:

bool arch_pmu__has_intel_topdown_events(void)

{
    static bool has_topdown_events = pmu_have_event("cpu", "slots") ||

pmu_have_event("cpu_core", "slots");

    return has_topdown_events;
}

bool arch_pmu__supports_intel_topdown_events(const char *pmu_name)

{
    if (!pmu_name)
      return false;
    return arch_pmu__has_intel_topdown_events() && (!strncmp(pmu_name,

"cpu") || !strncmp(pmu_name, "cpu_core"));
}

bool arch_evsel__is_intel_topdown_event(struct evsel *evsel)

{
    if (!arch_pmu__supports_intel_topdown_events(evsel->pmu))
      return false;

    return strcasestr(evsel->name, "slots") || strcasestr(evsel->name, "topdown");
}

This then gives us:

bool arch_evsel__must_be_in_group(const struct evsel *evsel)

{
    return  arch_evsel__is_intel_topdown_event(evsel);
}

These functions can then be reused for the arch_evlist topdown code, etc. What I don't see in these functions is use of any hybrid abstraction and so it isn't clear to me how with hybrid something like this would be plumbed in.

Thanks,
Ian

Quoted 6 lines
tools/perf/arch/x86/util/evsel.c +0 -1
                  return false;
 
          return evsel->name &&
--

2.35.1

#6 lore raw

PATCH 2/4 perf stat: Always keep perf metrics topdown events in a group

In reply to [email protected]

From: Kan Liang <[email protected]>

If any member in a group has a different cpu mask than the other members, the current perf stat disables group. when the perf metrics topdown events are part of the group, the below <not supported> error will be triggered.

$ perf stat -e "{slots,topdown-retiring,uncore_imc_free_running_0/dclk/}" -a sleep 1
WARNING: grouped events cpus do not match, disabling group:
  anon group { slots, topdown-retiring, uncore_imc_free_running_0/dclk/ }

Performance counter stats for 'system wide':

       141,465,174      slots
   <not supported>      topdown-retiring
     1,605,330,334      uncore_imc_free_running_0/dclk/

The perf metrics topdown events must always be grouped with a slots event as leader.

With the patch, the topdown events aren't broken from the group for the splitting.

$ perf stat -e "{slots,topdown-retiring,uncore_imc_free_running_0/dclk/}" -a sleep 1
WARNING: grouped events cpus do not match, disabling group:
  anon group { slots, topdown-retiring, uncore_imc_free_running_0/dclk/ }

Performance counter stats for 'system wide':

       346,110,588      slots
       124,608,256      topdown-retiring
     1,606,869,976      uncore_imc_free_running_0/dclk/
       1.003877592 seconds time elapsed
Fixes a9a1790247bd ("perf stat: Ensure group is defined on top of the same cpu mask")
---
 tools/perf/builtin-stat.c | 7 +++++--
 1 file changed, 5 insertions(+), 2 deletions(-)
tools/perf/builtin-stat.c +5 -2
index a96f106dc93a..af2248868a4f 100644
@@ -272,8 +272,11 @@ static void evlist__check_cpu_maps(struct evlist *evlist)
 		}
 
 		for_each_group_evsel(pos, leader) {
-			evsel__set_leader(pos, pos);
-			pos->core.nr_members = 0;
+			if (!evsel__must_be_in_group(pos) && pos != leader) {
+				evsel__set_leader(pos, pos);
+				pos->core.nr_members = 0;
+				leader->core.nr_members--;
+			}
 		}
 		evsel->core.leader->nr_members = 0;
 	}
Signature · 1 line
2.35.1
reviewer
#7 lore raw

Re: [PATCH 2/4] perf stat: Always keep perf metrics topdown events in a group

Quoted 28 lines · On Fri, May 13, 2022 at 8:16 AM <[email protected]> wrote:

From: Kan Liang <[email protected]>

If any member in a group has a different cpu mask than the other members, the current perf stat disables group. when the perf metrics topdown events are part of the group, the below <not supported> error will be triggered.

$ perf stat -e "{slots,topdown-retiring,uncore_imc_free_running_0/dclk/}" -a sleep 1
WARNING: grouped events cpus do not match, disabling group:
  anon group { slots, topdown-retiring, uncore_imc_free_running_0/dclk/ }

Performance counter stats for 'system wide':

       141,465,174      slots
   <not supported>      topdown-retiring
     1,605,330,334      uncore_imc_free_running_0/dclk/

The perf metrics topdown events must always be grouped with a slots event as leader.

With the patch, the topdown events aren't broken from the group for the splitting.

$ perf stat -e "{slots,topdown-retiring,uncore_imc_free_running_0/dclk/}" -a sleep 1
WARNING: grouped events cpus do not match, disabling group:
  anon group { slots, topdown-retiring, uncore_imc_free_running_0/dclk/ }
Quoted 7 lines

Performance counter stats for 'system wide':

       346,110,588      slots
       124,608,256      topdown-retiring
     1,606,869,976      uncore_imc_free_running_0/dclk/
       1.003877592 seconds time elapsed

Nice! This is based on:
https://lore.kernel.org/lkml/[email protected]/
You may end up with a group with the leader having a group count of 1 (itself). I explicitly zeroed that in the change above, but this may be unnecessary. Maybe we should move this code to helper functions for sharing and consistency on what the leader count should be.

Thanks,
Ian

Quoted 27 lines
Fixes a9a1790247bd ("perf stat: Ensure group is defined on top of the same cpu mask")
---
 tools/perf/builtin-stat.c | 7 +++++--
 1 file changed, 5 insertions(+), 2 deletions(-)
tools/perf/builtin-stat.c +5 -3
index a96f106dc93a..af2248868a4f 100644
@@ -272,8 +272,11 @@ static void evlist__check_cpu_maps(struct evlist *evlist)
                }
 
                for_each_group_evsel(pos, leader) {
-                       evsel__set_leader(pos, pos);
-                       pos->core.nr_members = 0;
+                       if (!evsel__must_be_in_group(pos) && pos != leader) {
+                               evsel__set_leader(pos, pos);
+                               pos->core.nr_members = 0;
+                               leader->core.nr_members--;
+                       }
                }
                evsel->core.leader->nr_members = 0;
        }
--

2.35.1

author
#8 lore raw

Re: [PATCH 2/4] perf stat: Always keep perf metrics topdown events in a group

Quoted 35 lines · Ian Rogers
Quoted 34 lines · On Fri, May 13, 2022 at 8:16 AM <[email protected]> wrote:

From: Kan Liang <[email protected]>

If any member in a group has a different cpu mask than the other members, the current perf stat disables group. when the perf metrics topdown events are part of the group, the below <not supported> error will be triggered.

$ perf stat -e "{slots,topdown-retiring,uncore_imc_free_running_0/dclk/}" -a sleep 1
WARNING: grouped events cpus do not match, disabling group:
   anon group { slots, topdown-retiring, uncore_imc_free_running_0/dclk/ }

Performance counter stats for 'system wide':

        141,465,174      slots
    <not supported>      topdown-retiring
      1,605,330,334      uncore_imc_free_running_0/dclk/

The perf metrics topdown events must always be grouped with a slots event as leader.

With the patch, the topdown events aren't broken from the group for the splitting.

$ perf stat -e "{slots,topdown-retiring,uncore_imc_free_running_0/dclk/}" -a sleep 1
WARNING: grouped events cpus do not match, disabling group:
   anon group { slots, topdown-retiring, uncore_imc_free_running_0/dclk/ }

Performance counter stats for 'system wide':

        346,110,588      slots
        124,608,256      topdown-retiring
      1,606,869,976      uncore_imc_free_running_0/dclk/
Quoted 9 lines
Quoted 1 line
        1.003877592 seconds time elapsed

Nice! This is based on:
https://lore.kernel.org/lkml/[email protected]/
You may end up with a group with the leader having a group count of 1 (itself). I explicitly zeroed that in the change above, but this may be unnecessary. Maybe we should move this code to helper functions for sharing and consistency on what the leader count should be.

I think the current code has already did evsel->core.leader->nr_members = 0; for the leader. So I don't change it.

Yes, a helper function seems reasonable. I will add it in V2.

Thanks,
Kan

Quoted 30 lines

Thanks,
Ian

Quoted 27 lines
Fixes a9a1790247bd ("perf stat: Ensure group is defined on top of the same cpu mask")
---
  tools/perf/builtin-stat.c | 7 +++++--
  1 file changed, 5 insertions(+), 2 deletions(-)
tools/perf/builtin-stat.c +5 -3
index a96f106dc93a..af2248868a4f 100644
@@ -272,8 +272,11 @@ static void evlist__check_cpu_maps(struct evlist *evlist)
                 }
 
                 for_each_group_evsel(pos, leader) {
-                       evsel__set_leader(pos, pos);
-                       pos->core.nr_members = 0;
+                       if (!evsel__must_be_in_group(pos) && pos != leader) {
+                               evsel__set_leader(pos, pos);
+                               pos->core.nr_members = 0;
+                               leader->core.nr_members--;
+                       }
                 }
                 evsel->core.leader->nr_members = 0;
         }
--

2.35.1

#9 lore raw

PATCH 3/4 perf parse-events: Support different format of the topdown event name

In reply to [email protected]

From: Kan Liang <[email protected]>

The evsel->name may have a different format for a topdown event, a pure topdown name (e.g., topdown-fe-bound), or a PMU name + a topdown name (e.g., cpu/topdown-fe-bound/). The cpu/topdown-fe-bound/ kind format isn't supported by the arch_evlist__leader(). This format is a very common format for a hybrid platform, which requires specifying the PMU name for each event.

Without the patch,

$perf stat -e '{instructions,slots,cpu/topdown-fe-bound/}' -a sleep 1

Performance counter stats for 'system wide':

     <not counted>      instructions
     <not counted>      slots
   <not supported>      cpu/topdown-fe-bound/
       1.003482041 seconds time elapsed

Some events weren't counted. Try disabling the NMI watchdog:
echo 0 > /proc/sys/kernel/nmi_watchdog
perf stat ...
echo 1 > /proc/sys/kernel/nmi_watchdog
The events in group usually have to be from the same PMU. Try reorganizing the group.

With the patch,

perf stat -e '{instructions,slots,cpu/topdown-fe-bound/}' -a sleep 1

Performance counter stats for 'system wide':

       157,383,996      slots
        25,011,711      instructions
        27,441,686      cpu/topdown-fe-bound/
       1.003530890 seconds time elapsed
Fixes bc355822f0d9 ("perf parse-events: Move slots only with topdown")
---
 tools/perf/arch/x86/util/evlist.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
tools/perf/arch/x86/util/evlist.c +1 -1
index cfc208d71f00..75564a7df15b 100644
@@ -36,7 +36,7 @@ struct evsel *arch_evlist__leader(struct list_head *list)
 				if (slots == first)
 					return first;
 			}
-			if (!strncasecmp(evsel->name, "topdown", 7))
+			if (strcasestr(evsel->name, "topdown"))
 				has_topdown = true;
 			if (slots && has_topdown)
 				return slots;
Signature · 1 line
2.35.1
reviewer Reviewed-by on v1 3/4
#10 lore raw

Re: [PATCH 3/4] perf parse-events: Support different format of the topdown event name

Quoted 35 lines · On Fri, May 13, 2022 at 8:16 AM <[email protected]> wrote:

From: Kan Liang <[email protected]>

The evsel->name may have a different format for a topdown event, a pure topdown name (e.g., topdown-fe-bound), or a PMU name + a topdown name (e.g., cpu/topdown-fe-bound/). The cpu/topdown-fe-bound/ kind format isn't supported by the arch_evlist__leader(). This format is a very common format for a hybrid platform, which requires specifying the PMU name for each event.

Without the patch,

$perf stat -e '{instructions,slots,cpu/topdown-fe-bound/}' -a sleep 1

Performance counter stats for 'system wide':

     <not counted>      instructions
     <not counted>      slots
   <not supported>      cpu/topdown-fe-bound/
       1.003482041 seconds time elapsed

Some events weren't counted. Try disabling the NMI watchdog:
echo 0 > /proc/sys/kernel/nmi_watchdog
perf stat ...
echo 1 > /proc/sys/kernel/nmi_watchdog
The events in group usually have to be from the same PMU. Try reorganizing the group.

With the patch,

perf stat -e '{instructions,slots,cpu/topdown-fe-bound/}' -a sleep 1

Performance counter stats for 'system wide':

Quoted 8 lines
       157,383,996      slots
        25,011,711      instructions
        27,441,686      cpu/topdown-fe-bound/
       1.003530890 seconds time elapsed
Fixes bc355822f0d9 ("perf parse-events: Move slots only with topdown")

Thanks,
Ian

Quoted 20 lines
---
 tools/perf/arch/x86/util/evlist.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
tools/perf/arch/x86/util/evlist.c +1 -2
index cfc208d71f00..75564a7df15b 100644
@@ -36,7 +36,7 @@ struct evsel *arch_evlist__leader(struct list_head *list)
                                if (slots == first)
                                        return first;
                        }
-                       if (!strncasecmp(evsel->name, "topdown", 7))
+                       if (strcasestr(evsel->name, "topdown"))
                                has_topdown = true;
                        if (slots && has_topdown)
                                return slots;
--

2.35.1

#11 lore raw

PATCH 4/4 perf parse-events: Move slots event for the hybrid platform too

In reply to [email protected]

From: Kan Liang <[email protected]>

The commit 94dbfd6781a0 ("perf parse-events: Architecture specific leader override") introduced a feature to reorder the slots event to fulfill the restriction of the perf metrics topdown group. But the feature doesn't work on the hybrid machine.

$perf stat -e "{cpu_core/instructions/,cpu_core/slots/,cpu_core/topdown-retiring/}" -a sleep 1

Performance counter stats for 'system wide':

     <not counted>      cpu_core/instructions/
     <not counted>      cpu_core/slots/
   <not supported>      cpu_core/topdown-retiring/
       1.002871801 seconds time elapsed

A hybrid platform has a different PMU name for the core PMUs, while current perf hard code the PMU name "cpu".

Introduce a new function to check whether the system supports the perf metrics feature. The result is cached for the future usage.

For X86, the core PMU name always has "cpu" prefix.

With the patch,

$perf stat -e "{cpu_core/instructions/,cpu_core/slots/,cpu_core/topdown-retiring/}" -a sleep 1

Performance counter stats for 'system wide':

        76,337,010      cpu_core/slots/
        10,416,809      cpu_core/instructions/
        11,692,372      cpu_core/topdown-retiring/
       1.002805453 seconds time elapsed
---
 tools/perf/arch/x86/util/evlist.c  |  5 +++--
 tools/perf/arch/x86/util/topdown.c | 18 ++++++++++++++++++
 tools/perf/arch/x86/util/topdown.h |  7 +++++++
 3 files changed, 28 insertions(+), 2 deletions(-)
 create mode 100644 tools/perf/arch/x86/util/topdown.h
tools/perf/arch/x86/util/evlist.c +3 -2
index 75564a7df15b..68f681ad54c1 100644
@@ -3,6 +3,7 @@
 #include "util/pmu.h"
 #include "util/evlist.h"
 #include "util/parse-events.h"
+#include "topdown.h"
 
 #define TOPDOWN_L1_EVENTS	"{slots,topdown-retiring,topdown-bad-spec,topdown-fe-bound,topdown-be-bound}"
 #define TOPDOWN_L2_EVENTS	"{slots,topdown-retiring,topdown-bad-spec,topdown-fe-bound,topdown-be-bound,topdown-heavy-ops,topdown-br-mispredict,topdown-fetch-lat,topdown-mem-bound}"
@@ -25,12 +26,12 @@ struct evsel *arch_evlist__leader(struct list_head *list)
 
 	first = list_first_entry(list, struct evsel, core.node);
 
-	if (!pmu_have_event("cpu", "slots"))
+	if (!topdown_sys_has_perf_metrics())
 		return first;
 
 	/* If there is a slots event and a topdown event then the slots event comes first. */
 	__evlist__for_each_entry(list, evsel) {
-		if (evsel->pmu_name && !strcmp(evsel->pmu_name, "cpu") && evsel->name) {
+		if (evsel->pmu_name && !strncmp(evsel->pmu_name, "cpu", 3) && evsel->name) {
 			if (strcasestr(evsel->name, "slots")) {
 				slots = evsel;
 				if (slots == first)
tools/perf/arch/x86/util/topdown.c +18 -0
index 2f3d96aa92a5..95b9fdef59ab 100644
@@ -3,6 +3,24 @@
 #include "api/fs/fs.h"
 #include "util/pmu.h"
 #include "util/topdown.h"
+#include "topdown.h"
+
+bool topdown_sys_has_perf_metrics(void)
+{
+	static bool has_perf_metrics;
+	static bool cached;
+	struct perf_pmu *pmu;
+
+	if (cached)
+		return has_perf_metrics;
+
+	pmu = perf_pmu__find_by_type(PERF_TYPE_RAW);
+	if (pmu && pmu_have_event(pmu->name, "slots"))
+		has_perf_metrics = true;
+
+	cached = true;
+	return has_perf_metrics;
+}
 
 /*
  * Check whether we can use a group for top down.
tools/perf/arch/x86/util/topdown.h new file +7 -0
new file mode 100644
index 000000000000..46bf9273e572
@@ -0,0 +1,7 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+#ifndef _TOPDOWN_H
+#define _TOPDOWN_H 1
+
+bool topdown_sys_has_perf_metrics(void);
+
+#endif
Signature · 1 line
2.35.1
reviewer
#12 lore raw

Re: [PATCH 4/4] perf parse-events: Move slots event for the hybrid platform too

Quoted 82 lines · On Fri, May 13, 2022 at 8:16 AM <[email protected]> wrote:

From: Kan Liang <[email protected]>

The commit 94dbfd6781a0 ("perf parse-events: Architecture specific leader override") introduced a feature to reorder the slots event to fulfill the restriction of the perf metrics topdown group. But the feature doesn't work on the hybrid machine.

$perf stat -e "{cpu_core/instructions/,cpu_core/slots/,cpu_core/topdown-retiring/}" -a sleep 1

Performance counter stats for 'system wide':

     <not counted>      cpu_core/instructions/
     <not counted>      cpu_core/slots/
   <not supported>      cpu_core/topdown-retiring/
       1.002871801 seconds time elapsed

A hybrid platform has a different PMU name for the core PMUs, while current perf hard code the PMU name "cpu".

Introduce a new function to check whether the system supports the perf metrics feature. The result is cached for the future usage.

For X86, the core PMU name always has "cpu" prefix.

With the patch,

$perf stat -e "{cpu_core/instructions/,cpu_core/slots/,cpu_core/topdown-retiring/}" -a sleep 1

Performance counter stats for 'system wide':

        76,337,010      cpu_core/slots/
        10,416,809      cpu_core/instructions/
        11,692,372      cpu_core/topdown-retiring/
       1.002805453 seconds time elapsed
---
 tools/perf/arch/x86/util/evlist.c  |  5 +++--
 tools/perf/arch/x86/util/topdown.c | 18 ++++++++++++++++++
 tools/perf/arch/x86/util/topdown.h |  7 +++++++
 3 files changed, 28 insertions(+), 2 deletions(-)
 create mode 100644 tools/perf/arch/x86/util/topdown.h
tools/perf/arch/x86/util/evlist.c +3 -2
index 75564a7df15b..68f681ad54c1 100644
@@ -3,6 +3,7 @@
 #include "util/pmu.h"
 #include "util/evlist.h"
 #include "util/parse-events.h"
+#include "topdown.h"
 
 #define TOPDOWN_L1_EVENTS      "{slots,topdown-retiring,topdown-bad-spec,topdown-fe-bound,topdown-be-bound}"
 #define TOPDOWN_L2_EVENTS      "{slots,topdown-retiring,topdown-bad-spec,topdown-fe-bound,topdown-be-bound,topdown-heavy-ops,topdown-br-mispredict,topdown-fetch-lat,topdown-mem-bound}"
@@ -25,12 +26,12 @@ struct evsel *arch_evlist__leader(struct list_head *list)
 
        first = list_first_entry(list, struct evsel, core.node);
 
-       if (!pmu_have_event("cpu", "slots"))
+       if (!topdown_sys_has_perf_metrics())
                return first;
 
        /* If there is a slots event and a topdown event then the slots event comes first. */
        __evlist__for_each_entry(list, evsel) {
-               if (evsel->pmu_name && !strcmp(evsel->pmu_name, "cpu") && evsel->name) {
+               if (evsel->pmu_name && !strncmp(evsel->pmu_name, "cpu", 3) && evsel->name) {
                        if (strcasestr(evsel->name, "slots")) {
                                slots = evsel;
                                if (slots == first)
tools/perf/arch/x86/util/topdown.c +1 -0
index 2f3d96aa92a5..95b9fdef59ab 100644
@@ -3,6 +3,24 @@
 #include "api/fs/fs.h"
 #include "util/pmu.h"
 #include "util/topdown.h"
+#include "topdown.h"

tools/perf/arch/x86/util/topdown.c line 16

Quoted 10 lines
tools/perf/arch/x86/util/topdown.c +10 -0
+
+bool topdown_sys_has_perf_metrics(void)
+{
+       static bool has_perf_metrics;
+       static bool cached;
+       struct perf_pmu *pmu;
+
+       if (cached)
+               return has_perf_metrics;
+

Worth a comment here for the meaning of looking up PERF_TYPE_RAW.

Thanks,
Ian

Quoted 26 lines
tools/perf/arch/x86/util/topdown.c +7 -0
+       pmu = perf_pmu__find_by_type(PERF_TYPE_RAW);
+       if (pmu && pmu_have_event(pmu->name, "slots"))
+               has_perf_metrics = true;
+
+       cached = true;
+       return has_perf_metrics;
+}
 
 /*
  * Check whether we can use a group for top down.
tools/perf/arch/x86/util/topdown.h new file +7 -1
new file mode 100644
index 000000000000..46bf9273e572
@@ -0,0 +1,7 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+#ifndef _TOPDOWN_H
+#define _TOPDOWN_H 1
+
+bool topdown_sys_has_perf_metrics(void);
+
+#endif
--

2.35.1